German is one written language spread over several ways of speaking. A news presenter in Hamburg, a lecturer in Munich and a manager in Vienna all speak Hochdeutsch, but with different vowels, a different melody and words of their own: an Austrian says Jänner and Paradeiser where a German says Januar and Tomate. Speech of this kind, standard German with a regional accent, is what the model handles best, and the transcript comes out in ordinary written German whichever side of the border the speaker is from.
Swiss German is a different case. In Switzerland people speak a dialect, Schwiizerdütsch, in everyday life and write standard German, so there is no agreed way to put the dialect on paper. When a Swiss speaker talks in standard German, for example in a lecture or a presentation, the result is good. A conversation held in broad Zurich or Bernese dialect, or in strong Bavarian or Swabian, is transcribed into the nearest standard German words, and it needs more corrections, much as it would for a listener from northern Germany.
The transcript is written with ä, ö, ü and ß in place, not with ae, oe and ue, and nouns start with a capital letter as German spelling requires. That matters more than it seems: a text written with the wrong capitals or without umlauts has to be retyped before it can go into a thesis, a set of minutes or a subtitle file. Swiss spelling uses ss instead of ß, so a text for a Swiss audience may need a search and replace.
German builds long words by joining shorter ones, and speakers coin new ones as they go: Datenschutzbeauftragte, Projektabschlussbesprechung. Common compounds come out as one word, as they should. Rare or freshly made ones are sometimes split in two, and company names, product names and surnames such as Meier, Maier or Meyer cannot be decided by ear, so read those through once before the text is published.
Many German recordings here are university lectures and seminars, a Vorlesung of ninety minutes recorded at the back of the hall or exported from Zoom. Then come work meetings in Teams, interviews for a Bachelor or Master thesis, which in German-speaking universities usually have to be transcribed in full for the appendix, and podcasts and talks on YouTube that run well past an hour. A video file is accepted as it is; the speech is taken from it and the picture is ignored.
German at work is full of English: das Meeting, der Call, einen Termin canceln, das Deadline-Thema. Single English terms inside a German sentence are usually written correctly. Long stretches where speakers switch language every few sentences are harder, and a word may come out spelled by German ear. These are quick to fix in the editor that comes with a free account.
Without an account you get 10 transcriptions of up to 3 hours each, enough for a full lecture or a long meeting in one upload. The free allowance is 5 hours in total across those 10 recordings, and a free account keeps the full transcripts of what you transcribed as a guest. AI chat with the transcript, automatic summaries and export to PDF or Word belong to the paid plans.
The speech models run on our own servers. The recording is not passed to a third-party speech API, which matters when the file is a client call, a thesis interview or a meeting that was never meant to leave the room.